Bayesian estimation and testing in random-effects meta-analysis of rare binary events allowing for flexible group...

Area of Science:

• Medical Statistics
• Biostatistics
• Clinical Trial Analysis

Background:

• Rare binary events data are common in medical research, but individual studies often lack statistical power.
• Traditional meta-analysis methods can yield biased estimates for rare events and rely on restrictive directional assumptions.
• Combining evidence from multiple studies is crucial for robust conclusions in rare-event research.

Approach:

• Propose novel Bayesian procedures using a flexible random-effects model that does not assume a pre-specified direction of variability.
• Develop a Markov chain Monte Carlo algorithm with Pólya-Gamma augmentation for computational efficiency.
• Estimate and test overall treatment effects and inter-study heterogeneity in rare-event meta-analyses.

Key Points:

• The proposed Bayesian approach yields less biased and more stable estimates compared to existing methods.
• The flexible model accommodates various patterns of inter-study heterogeneity without directional constraints.
• Pólya-Gamma augmentation significantly enhances the computational efficiency of the algorithm.

Conclusions:

• The novel Bayesian procedures offer a more reliable and computationally efficient tool for meta-analysis of rare binary events data.
• This method addresses limitations of traditional approaches, improving the accuracy of treatment effect estimation in medical research.
• Demonstrated utility with real-world data from rosiglitazone and stomach ulcer studies.